Alexandre Prokoudine, who translated David’s interview in Russian, asked for some more precisions from David regarding using GREYCstoration to denoise a RAW image before dematricing. That is the approach that DXO 5 uses, apparently with very good results. Here is the extract:
AP: Can GREYCstoration actually be used before Bayer reconstruction already or with some (what exactly?) changes in the future?
DT: Concerning the RGB Bayer reconstruction, I already tried to adapt the PDE’s used in GREYCstoration to handle this case, but without much success. I guess GREYCstoration can be used as a post-process perhaps to remove some reconstruction artefacts, but I don’t see a serious way to use it before the reconstruction step.
Concerning the future changes, I have actually a lot of ideas and an on-going research work that could be integrated in GREYCstoration. I’m writing a research paper to describe all these techniques. Once this paper is accepted for publication in a conference, I plan to integrate those stuffs as soon as possible.
Since then, a new version of GREYCstoration ( 2.8 ) has landed; it has a new denoising method called “patch-based” which offers, according to David’s announcement “quite interesting denoising results when parameters are set correctly, sometimes better than with the traditional ‘-denoise’ method.” It is based on this research paper by David.
The 2.8 version of GREYCstoration offers a patch denoising option in the command line and in the Linux based Gimp plug-in. So far, my testing hasn’t proved ground-breaking; no significant enhancement compared to what can be achieved with “normal” denoising. But setting parameters correctly isn’t a very easy task so I may have missed something.